From Legal War to Licensed Generation: Suno, BMG, and the New Architecture of the Music Industry

Suno did not merge with BMG through a merger or acquisition.

On August 12, 2026, the two companies announced a global strategic alliance.

According to BMG’s official statement, the agreement establishes a framework covering:

BMG recorded music;

BMG publishing repertoire;

future AI-based music experiences;

compensation for participating artists and songwriters;

a voluntary participation model;

and the resolution of Suno’s prior use of BMG works and recordings. (bmg.com)

The final point is particularly important.

It means that the agreement is not only prospective — defining what Suno may use tomorrow — but also addresses the question of what the model may have used yesterday.

The financial terms of the agreement have not been publicly disclosed. Nor has the exact formula for distributing AI-related revenue among BMG, songwriters, artists, and Suno been made public. Any claim about specific royalty percentages would therefore be speculative at this stage.

BMG refers to “clear economics,” but the economics themselves have not been publicly described. (bmg.com)

BMG is far from being a small independent publisher.

The company is owned by Bertelsmann and is one of the world’s largest music companies. According to Bertelsmann, BMG represents more than three million songs and recordings. In 2025, its revenue was approximately €900 million, with adjusted operating EBITDA of €284 million. Digital business accounted for 71% of revenue. (bmg.com)

In addition, in April 2026, BMG and Concord announced plans to combine their businesses. The proposed combined company was described as having approximately $2.2 billion in pro forma annual revenue and more than $730 million in pro forma EBITDA. (bmg.com)

The arrival of BMG within the Suno ecosystem is therefore fundamentally different from an agreement involving a small catalog or a single artist.

This is already becoming a question of large-scale market formation for licensed AI training data.

At the same time, it would be incorrect to assume automatically that the entire Concord catalog is already covered by the Suno–BMG agreement. The public announcement of August 12 refers specifically to BMG repertoire. Until the combination is completed and additional terms are published, extending the agreement to Concord would be unjustified.

To understand the scale of the change, we need to return to 2024.

On June 24, 2024, record companies associated with Universal Music Group, Sony Music Entertainment, and Warner Music Group filed lawsuits against Suno and Udio through the RIAA.

The central allegation was that the AI companies had used enormous quantities of copyrighted sound recordings to train their models without permission from rights holders. (riaa.com)

At first, there was considerable opacity surrounding Suno’s training data.

But by August 2024, the company’s position had become much clearer.

In court filings, Suno acknowledged that building its model required analyzing tens of millions of recordings. The company did not deny the presence of copyrighted material, but argued that such use fell under the American legal doctrine of fair use. (musicbusinessworldwide.com)

This became the central philosophical and legal conflict of the first generation of generative AI.

Suno’s position was roughly that a machine learns musical patterns in a way comparable to a human being listening to existing music and learning how to create something new.

Rights holders took a fundamentally different view: an AI company was carrying out mass commercial copying of protected recordings in order to build technology that could then compete with those same works and performers.

U.S. law has still not produced one final universal answer to that question.

However, the direction of the debate is becoming increasingly visible.

In 2025, the U.S. Copyright Office published a major report on generative AI training.

Its position is much more nuanced than either “AI training is infringement” or “AI training is fair use.”

The Copyright Office emphasizes that the assessment depends on the specific material involved, the source of the data, the purpose of use, the design of the model, and the potential harm to existing markets.

At the same time, the Office makes an exceptionally important observation: large-scale commercial use of copyrighted works to generate expressive content capable of competing in the same markets — particularly where source material was obtained unlawfully — may extend beyond traditional concepts of fair use. (copyright.gov)

The Copyright Office also recommended not immediately creating a universal government licensing system, instead allowing voluntary licensing markets to develop first. (copyright.gov)

This is precisely why agreements such as Warner–Suno and BMG–Suno are so important.

The market is effectively beginning to build the mechanism envisioned by the Copyright Office:

AI company gains access to high-quality content → rights holder receives payment → contractual rules are created → litigation risk decreases → a licensing market for AI training emerges.

In other words, the dispute is gradually moving from the courtroom into the contract.

The turning point came in the autumn of 2025.

On October 29, Universal Music Group and Udio announced agreements providing for the settlement of litigation and the creation of a new AI service based on licensed and authorized music. (universalmusic.com)

On November 19, Warner Music Group reached a similar agreement with Udio. (wmg.com)

Then, on November 25, an event occurred that directly preceded the BMG–Suno agreement:

Warner Music Group and Suno settled their legal conflict and became partners.

Warner articulated three fundamental principles for its AI strategy:

models should be licensed;

the economic model should reflect the value of music;

artists and songwriters should be able to decide for themselves whether their names, images, likenesses, voices, and compositions may be used. (wmg.com)

Suno also committed to transitioning toward a new generation of licensed models.

More importantly, older models were expected to be gradually retired. (investors.wmg.com)

This marked a qualitatively different phase.

In June 2026, Suno raised more than $400 million in Series D funding at a valuation of approximately $5.4 billion. (suno.com)

This is a major signal.

We are no longer dealing with a small AI startup experimenting with music generation, but with a company possessing enough capital to:

license large catalogs;

fund litigation;

develop foundation music models;

build professional production tools;

create its own distribution and consumption ecosystem;

and compete for a new form of music consumption.

In June, Suno explicitly stated that it was preparing the first model built together with the music industry. (suno.com)

Then events began to unfold in rapid succession.

On July 10, IFPI, RIAA, IMPALA, WIN, A2IM, the Recording Academy, and other industry organizations introduced a common framework for labeling music.

The proposal draws a fundamental distinction between:

AI-Generated

and

AI-Assisted.

This terminology matters enormously.

The industry is effectively acknowledging that the issue is no longer simply “AI or no AI.”

Instead, there is a spectrum of technological participation in music creation. (ifpi.org)

On July 30, IFPI went further and introduced principles governing the eligibility of AI-related recordings for official charts.

Three core conditions were proposed:

the AI tool must be authorized and lawful;

the recording must remain predominantly the result of human creativity;

there must be no evidence of streaming or chart manipulation. (ifpi.org)

In practical terms, this creates a new institutional category:

not “AI is prohibited,” but “licensed AI can be part of normal music production.”

On August 6, 2026, Suno published a new declaration of principles.

The change in language is particularly revealing.

Instead of primarily emphasizing technological freedom and fair use, the company introduced a new formula:

Great Music is Made by People.

Suno pledged to introduce provenance mechanisms, watermarking, and fingerprinting, and to combat mass-generation abuse and unauthorized artist imitation. (suno.com)

On August 10, Suno announced another major change.

Beginning September 3, 2026, download limits would be introduced:

Free — up to 7 trial downloads in total;

Pro — 20 per month;

Premier — 60 per month;

while professional Suno Studio use under Premier would continue to include unlimited downloads. (suno.com)

At first glance, this looks like a pricing or subscription adjustment.

Strategically, however, it means much more.

Suno is gradually moving away from a system built around:

“generate an unlimited number of files and export them elsewhere.”

It is beginning to resemble:

a controlled music platform with regulated content export, provenance, licensing, and a professional production environment.

Two days later, the BMG agreement was announced.

That sequence is unlikely to be accidental.

One of the greatest problems in generative music is not that AI can create one good song.

The problem is scale economics.

A human being cannot physically release tens of thousands of new works every day.

An automated system can.

Deezer reported in July 2026 that it was receiving approximately 90,000 fully AI-generated tracks every day. At the June peak, such tracks represented more than half of all daily music uploads to the platform. (newsroom-deezer.com)

But here an extremely important contradiction appears.

Despite the enormous number of uploads, AI-generated tracks represented only around 1–3% of actual listening on Deezer.

At the same time, Deezer classified as much as 85% of streams on fully AI-generated tracks in 2025 as fraudulent and demonetized them. (newsroom-deezer.com)

This supports a fundamental conclusion:

an enormous volume of AI content does not yet translate into equivalent consumer demand for that content.

The main threat in the first phase is not that “people have stopped listening to humans.”

The more immediate threat is cheap industrial-scale content production, streaming fraud, and catalog pollution.

From this perspective, Suno’s download limits can reasonably be interpreted both as an anti-fraud measure and as part of a broader compromise with the established music industry.

On August 12, BMG became the first major rights holder after Warner to sign a full-scale agreement directly with Suno. At the time of the announcement, Universal Music Group and Sony Music were still pursuing litigation against Suno. (musicbusinessworldwide.com)

This is why the importance of BMG is greater than the size of its catalog alone.

Before BMG, Warner could still be interpreted as an exception.

After BMG, a pattern of behavior begins to emerge.

One major rights holder may be an experiment.

Two begin to form a market.

If a third, fourth, and fifth major catalog follow, licensed generative music will begin moving from exception toward industry standard.

A musical work is not a single legal object.

There are at least two fundamentally different layers of rights.

The first is the composition: melody, lyrics, and musical structure. This is where songwriter and publisher rights operate.

The second is the sound recording/master: the specific recording of a specific performance.

AI music may affect both layers.

That is why it is particularly significant that BMG’s official announcement explicitly refers to both recordings and music publishing repertoire. (bmg.com)

This potentially allows much more comprehensive licensing structures than simply granting permission to use master recordings.

For the future AI market, this architecture is critical.

There is also another layer that traditional music licensing has historically addressed only partially.

That layer includes:

voice;

performance style;

name;

image;

likeness;

artist persona.

Warner’s agreement with Suno explicitly stated that use of name, image, likeness, voice, and compositions should operate on an opt-in basis. (investors.wmg.com)

BMG’s public announcement is less specific regarding NILV, so it would be incorrect to assume automatically that the BMG agreement is identical to Warner’s.

But the direction of the market is clear.

A new licensable asset is emerging:

not only an artist’s song, but the artist’s digital creative identity itself.

And this could generate an enormous new market.

In the twentieth century, music economics was built around ownership:

buy the record.

In the early twenty-first century, the market shifted toward access:

get access through Spotify.

Generative AI may create a third stage:

interaction.

The listener no longer merely presses play.

They may interact with musical material.

For example:

create an alternative version;

change the arrangement;

make an authorized remix;

sing along;

change the genre context;

create a personalized composition based on a licensed musical universe associated with an artist.

That is why Suno and BMG refer not simply to “training models,” but to new music experiences and new ways for artists to connect with fans. (suno.com)

Economically, this could be extremely important.

Spotify monetizes listening.

A future AI platform may monetize creative interaction with music.

That is an entirely different product.

Today, the value of a composition on a platform is determined primarily by how many times it is played.

But if users pay for the ability to interact with a work, a different economic model emerges.

For example:

a single song could become the basis for thousands of user-created licensed remixes;

an artist’s voice could become the basis for authorized fan creations;

a composition could become the foundation for personalized variations;

a catalog could become the basis for training a specialized creative model.

Music therefore changes from a finished product into programmable intellectual property.

This may become one of the most fundamental economic effects of generative AI on music.

A catalog is no longer only something that can be listened to.

It becomes something with which users can legally create.

Imagine a hypothetical case.

A user creates a new work using a licensed AI system.

The model was trained on thousands of artists.

The user applies features associated with one licensed performer.

The underlying composition belongs to another songwriter.

The final recording includes original lyrics written by the user.

Who gets paid?

This is far more complicated than ordinary streaming royalties.

At minimum, value may need to be allocated among:

training data;

the specific licensed artist identity;

the underlying composition;

the user’s own creativity;

the model itself;

the platform.

This means that the future AI music economy will require a new attribution system.

The current public terms of the BMG–Suno agreement do not reveal how this system will work.

But technologically and economically, this may become one of the central questions of the next decade.

At first glance, generative AI should reduce the value of music: if music can be generated infinitely, scarcity disappears.

But there is a paradox.

Scarcity may disappear for generic music.

At the same time, scarcity may increase for:

recognizable works;

known artists;

unique voices;

historically significant catalogs;

high-quality human performance data.

In other words, AI may devalue “music in general” while increasing the value of music with provenance and identity.

This is one reason companies such as BMG continue acquiring catalogs aggressively while simultaneously developing AI strategies. In 2025, BMG reported that it had invested more than $1.5 billion in music-rights acquisitions since 2021 while also making AI one of its strategic business priorities. (bmg.com)

In this context, a catalog becomes something close to a high-quality licensed dataset combined with a cultural brand.

Areas where music is functional rather than artist-centered are particularly vulnerable:

background music;

production libraries;

jingles;

simple stock music;

generic soundtracks;

music for low-cost social content;

certain categories of advertising music.

In these markets, the client often does not care about the identity of the performer.

They care about function:

“we need two minutes of uplifting corporate pop.”

AI can automate this market very quickly.

A CISAC/PMP Strategy study projected that under certain regulatory assumptions, generative AI could place as much as 24% of music creators’ revenue at risk by 2028. Such figures should be understood as scenario-based estimates from a study commissioned by a rights-holder organization, not as established economic fact. (cisac.org)

But the direction of risk appears economically plausible:

generic music becomes cheaper; identifiable human artistry becomes more valuable.

This is where the situation becomes particularly interesting.

At first glance, major agreements involving BMG and Warner may seem relevant only to large rights holders.

But independent artists may ultimately have much to gain from the new structure.

If the market truly moves toward licensed AI, an independent artist may begin treating their own:

vocals;

stems;

guitar performances;

acoustic recordings;

melodies;

catalog;

performance data;

creative-style datasets

not only as material for releasing one recording, but as licensable creative capital.

A new professional role could emerge:

artist as model owner.

An artist does not necessarily need to sell all rights to an AI company.

They could grant narrowly defined permissions.

For example:

my voice may be used only within a specific platform;

fan remixes are allowed;

political or advertising use is prohibited;

training is allowed but voice cloning is prohibited;

commercial use is allowed only with revenue sharing.

If the technical and legal infrastructure is built correctly, this could significantly strengthen independent creators.

There is a serious risk that the licensed AI market will become accessible primarily to large corporations.

A major AI developer can negotiate with Warner or BMG.

A major rights holder can negotiate with Suno.

An individual songwriter in Prague, Berlin, or Kyiv cannot realistically negotiate separately with every AI company.

If the market becomes entirely contract-based, transaction costs could become enormous.

The U.S. Copyright Office addresses this problem when discussing collective licensing and extended collective licensing as possible future solutions if voluntary markets fail to reach certain categories of rights holders. (copyright.gov)

The key question of the next stage may therefore not simply be:

“Will AI pay?”

but:

“How will the money reach millions of smaller rights holders?”

The American system is moving largely through litigation and voluntary licensing markets.

Europe is simultaneously building a regulatory infrastructure.

Under the EU AI Act, providers of general-purpose AI models are required to maintain copyright policies and publish sufficiently detailed summaries of the content used for training. They must also respect rights reservations expressed by copyright holders. (digital-strategy.ec.europa.eu)

In July 2026, the European Commission even published a feasibility study examining an EU-level registry for text-and-data-mining opt-outs. (digital-strategy.ec.europa.eu)

In other words, an infrastructure is gradually emerging that may allow machines to determine:

this work is authorized;

this work is prohibited;

this work requires a license;

this artist has allowed a specific form of use.

This may ultimately matter more than the present-day quality of Suno’s music generation.

On July 31, 2026, a court in Munich ruled against Suno in a case brought by GEMA.

The court found copyright infringement, ordered Suno to disclose related profits, and awarded damages; Suno said it was considering an appeal. (reuters.com)

This occurred only twelve days before the BMG agreement was announced.

Legally, one event does not prove that the other happened because of it.

But the economic logic is clear.

The more jurisdictions create liability risk around unlicensed training, the more attractive it becomes to:

purchase legal certainty through licensing.

It would be incorrect to say:

“The music industry has reached an agreement with Suno.”

It has not.

Warner reached an agreement.

BMG reached an agreement.

Universal Music Group and Sony, at the time of the BMG announcement, remained in litigation against Suno. (musicbusinessworldwide.com)

On August 17, 2026 — only five days after the BMG announcement — Round Hill Music filed a new lawsuit against Suno involving alleged use of hundreds of song lyrics, with claims potentially extending to more than 10,000 works. (reuters.com)

Two realities therefore exist simultaneously:

a licensed AI music economy

and

a litigation economy surrounding unlicensed AI training.

They are developing in parallel.

BMG’s reference to resolving prior use has a problematic side.

If this practice becomes normalized, AI companies may receive a dangerous economic signal:

build the model;

use the content;

grow rapidly;

raise investment;

and then, once bargaining power has increased substantially, negotiate a retroactive settlement.

Such a system could create moral hazard.

A smaller startup that licenses data honestly from the beginning would incur much higher costs than a competitor that first trained on the open internet.

The future system will therefore need to answer another difficult question:

does the possibility of later legalization create a competitive advantage for initial infringement?

This is one of the most difficult issues in the emerging market.

There is an opposing perspective.

If every owner of every recording spends decades suing every AI developer, no viable market will emerge.

Settlement converts legal uncertainty into an economic system.

The rights holder receives compensation.

The AI company receives legal access.

Artists receive opt-in mechanisms.

Users receive new tools.

Instead of prohibition, a market emerges.

Historically, the music industry has often followed this pattern after major technological conflicts.

The most important change may ultimately be terminological.

Today, the phrase “AI music” is used to describe radically different things:

a fully automatically generated song with little or no human involvement;

a human-written song with AI arrangement;

AI restoration;

AI stem separation;

synthetic instruments;

AI-assisted mixing;

AI vocal transformation;

generated orchestration;

a licensed artist model;

mass-produced AI-generated commercial spam.

Technologically, these are very different.

Legally, they are different.

Culturally, they are different as well.

The emergence of the categories AI-Assisted and AI-Generated may therefore be only the first step toward a much more sophisticated classification of musical works. (ifpi.org)

None of these scenarios should be understood as a prediction. They are better interpreted as four possible directions that may coexist.

The most likely outcome may be a combination of these scenarios rather than any single one.

Until now, competition among AI music companies has focused primarily on:

generation quality;

speed;

control;

vocals;

track length;

stem quality.

Now another variable is emerging:

the legal provenance of the model.

A future model may be marketed not only as:

“we have the best sound,”

but also as:

“our model is trained on licensed material, includes provenance infrastructure, and is compatible with the commercial music industry.”

For professional artists, the second claim may sometimes matter more than the first.

Warner owns a massive catalog.

BMG owns a massive catalog.

If Suno gains lawful access to large, structured, high-quality music datasets, this creates not only a legal advantage.

It may also create a technological advantage.

Well-labeled professional masters, stems, compositions, and metadata may be significantly more valuable for model training than randomly scraped internet files.

A licensing agreement therefore becomes simultaneously:

a legal contract;

a content deal;

a dataset acquisition strategy;

an R&D partnership.

This creates a potential moat for Suno that smaller competitors may find extremely difficult to reproduce.

There is an obvious economic problem.

If training the best model requires hundreds of millions of dollars in investment, plus licenses from Warner, BMG, Universal, Sony, collecting societies, and thousands of publishers, the market becomes highly capital-intensive.

In such a world, the advantage goes to a handful of corporations.

The result may resemble the streaming market:

millions of creators;

but only a few infrastructure platforms.

AI, originally perceived as a tool for democratizing creativity, may therefore become highly centralized at the infrastructure level.

Generative AI makes the technical production of music dramatically cheaper.

But precisely because of that, human elements may become more valuable.

When technically perfect instrumental performance can be generated in seconds, technical execution is no longer the rarest resource.

The rare resource becomes:

personal history;

biography;

emotional authenticity;

recognizable human phrasing;

cultural context;

live performance;

the relationship between artist and audience;

verifiable human provenance.

AI may therefore shift economic scarcity:

from the production of sound to the provenance of meaning.

DSPs will inevitably become a third major participant in this transformation.

In the future, they may need to know far more than ISRC, songwriter, and label.

They may increasingly need information such as:

was generative AI used?

is the recording AI-assisted or AI-generated?

which AI system was used?

is that system licensed?

does provenance exist?

was voice or personality usage authorized?

does the track violate the AI platform’s terms?

is the track associated with streaming fraud?

IFPI is already moving in this direction through its AI eligibility principles for charts. (ifpi.org)

Suno, for its part, has announced watermarking and fingerprinting mechanisms. (suno.com)

These processes are beginning to converge.

Who made the music?

What material was the system trained on?

What did the human do?

What did the model do?

What rights were obtained?

Who owns the voice?

Who gets paid?

Can these claims be verified?

The answers to these questions will increasingly become part of the music product itself.

That is why provenance — origin and traceability — may become one of the most important technologies in the future music industry.

For the traditional music business, generative AI is simultaneously a threat and a new source of revenue.

The threat lies in the automation of production music, content oversupply, streaming fraud, replacement of certain commissioned-music markets, and the potential decline in the cost of technical production.

The opportunity lies in transforming large catalogs into a new category of licensable assets.

A catalog may generate revenue through:

streaming;

sync;

performance;

mechanical rights;

AI training;

interactive experiences;

authorized remixes;

digital artist identity.

Generative AI may therefore create a new layer of the music rights economy, rather than functioning merely as another production tool.

For AI music, the BMG–Suno agreement can be interpreted as one of the transition points between the first and second phases of the market.

Phase One

Technology first.

The main goal is to prove that a machine can create music at all.

Training data is opaque.

The legal model is secondary.

The primary benchmark is generation quality.

Phase Two

Institution first.

Model quality is already sufficiently high.

Now the central question becomes:

can this technology be integrated into the real economy?

Can it be used professionally?

Can music created with it be released commercially?

Can provenance be demonstrated?

Who gets paid?

Can permission be obtained?

Can artists participate directly?

The market is now entering this second phase.

An improvement from Suno v5 to v6 or v7 is a technological event.

A change in the rules of ownership and licensing is an institutional event.

Technological versions become obsolete quickly.

Institutional models can survive for decades.

Spotify is not historically important merely because it once had a better audio player.

It matters because it helped establish a specific relationship:

listener → platform → rights holder → royalty.

Generative AI is now searching for its own formula.

It may eventually look something like this:

creator/fan → AI platform → licensed model → rights attribution → creator/rightsholder compensation.

If such a chain becomes institutionalized, the BMG–Suno agreement may be remembered as one of the early agreements on which that architecture was built.

The weakest point of all the announced agreements is the lack of public information about how the money will actually be distributed.

We know there will be compensation.

We know there is opt-in.

We know licensed models are being developed.

But we know almost nothing about the mathematics.

Does the artist receive:

a fixed fee?

a percentage of subscription revenue?

payment for each use?

payment for training?

payment for each generation?

payment for remixing?

a share of revenue generated by the entire model?

How is value distributed among thousands of works used during training?

This question is fundamental.

Because “licensed AI” does not automatically mean a fair AI economy.

The true revolution will happen only when not only the provenance of training data, but also the provenance of money becomes transparent.

The Suno–BMG agreement does not mark the end of the conflict between AI and the music industry.

Rather, it demonstrates that the conflict has entered a new phase.

In 2024, the question was:

“Does AI have the right to train on our music?”

By 2026, the question is increasingly becoming:

“If AI uses our music, under what rules, with whose permission, and who receives the value created?”

That is a profound conceptual shift.

BMG is effectively acknowledging that generative music will not disappear.

Suno is effectively acknowledging that rights holders cannot be ignored indefinitely.

Both sides are beginning to build an economic system around the technology.

For the traditional music industry, this creates a new licensing market and new ways to monetize catalog IP.

For AI music, it marks the end of an era in which technical quality alone was sufficient as a competitive advantage.

The next generation of AI music will increasingly be evaluated according to four dimensions:

quality;

control;

legality;

traceability of provenance.

For artists themselves, perhaps the most interesting possibility is emerging here.

In the past, a musician sold a recording.

Later, musicians sold access to that recording.

In the future, artists may be able to license the very possibility of creative interaction with their musical identity.

That is where one of the largest potential markets may exist — a market that barely exists today.

The BMG–Suno agreement should therefore not be understood merely as news that “another label has allowed AI to use music.”

Its significance is much deeper.

The music industry is beginning to transform generative AI from an external disruptor of its business model into a licensed internal layer of the music economy.

If this model takes hold, the central debate of the next decade will no longer be about whether AI-generated music should exist.

It will be about who owns the value it creates.

Official BMG–Suno announcement: BMG — Global Strategic Alliance with Suno

Suno’s statement on the agreement: Suno — Landmark Global Partnership with BMG

Earlier Suno–Warner agreement: Warner Music Group — Suno Partnership

Suno’s new rules and download limits: Suno — Terms and Download Policy Changes

Suno’s responsible AI principles: Suno — Building the Future of Music Responsibly

IFPI rules for AI music and charts: IFPI — AI Recordings and Official Charts

Deezer statistics on AI music: Deezer — AI Music Exceeds 50% of Daily Uploads

EU regulation of GPAI: European Commission — General-Purpose AI Obligations